摘要
针对港口集装箱自动装卸问题,结合双目视觉和深度学习技术,设计了一种基于Faster R-CNN(faster regions with convolutional neural network)模型的集装箱三维识别定位方法。首先利用双目摄像头采集装箱图像,用于训练Faster R-CNN模型,然后使用该模型检测图像中的集装箱目标,对识别出的目标添加矩形框,并提取其中心点图像坐标,接着通过对双目摄像头进行标定和匹配,获取集装箱矩形框中图像坐标点的深度,实现对集装箱的三维定位,最后将集装箱的三维坐标转换到轮胎吊吊具坐标系下,获得所有集装箱目标中心和吊具中心距离。实验结果表明,系统运行速度可以达到30 fps,平均定位误差在5 mm以内,系统可以有效解决集装箱三维实时识别和定位问题,提升港口集装箱自动化装卸能力。
In view of the problems of automatic container operation,and combining binocular vision with deep learning,a container 3 D recognition and location method based on Faster R-CNN model is presented.The paper firstly uses the image information of the containers acquired by binocular camera to train Faster R-CNN model,then detects the targets according to the model for getting the rectangular box of the container and its center coordinate of frame.In order to get the depth of the containers to realize the three-dimensional positioning,the binocular depth camera has been calibrated and matched.Finally,the coordinates are converted into the coordinate system of the tire hanging spreader to calculate the distance between all the containers center and spreader center.The results show that the system can reach 30 fps,and the average position of error is within 5 mm.The system can effectively solve the three-dimensional real-time recognition and location problem of container,and improve the automatic operation level for handing the containers at port.
作者
丁小虎
陈宁
DING Xiaohu;CHEN Ning(School of Mechanical and Energy Engineering,Jimei University,Xiamen 361021,China)
出处
《集美大学学报（自然科学版）》
CAS
2019年第4期290-298,共9页
Journal of Jimei University：Natural Science
基金
福建省科技厅资助项目(JK214024)
国家海洋局(省海洋渔厅)资助项目(2014FJPT03)
福建省自然科学基金资助项目(2016J01755)
关键词
集装箱
双目视觉
深度学习
三维定位
container
binocular vision
deep learning
three-dimensional positioning
